TechReport's 8 PSU roundup.

[FreiDOg]

Tech Report has a nice review up of alot of the 'hot ticket' supplies, plus a few buget entries.

Perhaps not the most spectacular all around review (I'm particuarly disspaointed Thermaltake and PC P&C opted not to participate), but check out the AC ripple tests. While all (expect one) are well within spec in all the tests, a fanless 300W supply from 'Silverstone' shows up Enermax and NeoPower. (Silverstone is the most expensive supply in there, but hey, I was still surprised).

 

[potroast]

seeing PC P&C would have been nice, but TT can kiss my ass... they haven't ever made a decent product. For how loud their crap is you'd think it could cool a nuclear reactor -- but it can barely cool a stock rig.

 

[xonik]

Power supplies aren't meant to cool your whole system anymore. The fact that some of their power supplies are loud is neither a testament to the entire lineup nor a testament to the output quality and efficiency. After some time with their power supplies, you would probably find that while flashy, they are decent quality power supplies that compare favorably with the likes of Antec, Fortron, and Enermax.

 

[AACI_CIO]

What struck me is the UltraX Connect was, well, mediocre. I have seen a lot of people say the UltraX is great, but that review (at least for that unit) shows some potential doubts.

Also, the NeoPower 12v was 11.66? That is terrible. Moreover, why didn't they just up the volts on the OCZ? That's what the voltage adjusts are there to do!

 

[xonik]

In defense of the Antec unit, 11.66 volts is within spec. Also, a single measurement of 11.66 volts is statistically insignificant.

 

[potroast]

yeah, well count my vote that antec sucks. I've had WAY too many problems with them. 11.6V? yeah, it's within spec -- but being a great PSU doesn't mean you're within spec, but means you're WELL within spec... and consistant regardless of load.

Sorry, but antec is considered by me as a "good" power supply that doesn't belong in a "great" system... You want something better than the generic pos? Sure, get antec -- it's good... But if you're an enthusiast that wants the best, or needs something to power a kick-ass rig, then look elsewhere.

 

[MaMMa]

I got my first antec neo 480 and it was doa. 2 weeks ago. still waiting on my new one

 

[AACI_CIO]

In defense of the Antec unit, 11.66 volts is within spec. Also, a single measurement of 11.66 volts is statistically insignificant.

If that is a point measurement, then this WHOLE REPORT is statistically INVALID. Without replication, no error rate can be established and thus no statistical inference. If 11.66 is in fact a mean from a replicated experiment, then the measurement is, at least for that unit, statistically valid and likely significant in an F-Test.

However, to truly measure any of the units, you would need replicates of the units themselves. In fact, to be quite certain (within a single deviation), you would want ~39 units of each type if you had 5 factors of interest (read: ripple, volts, etc.) with an alpha of 0.05 and beta of 0.05 (see "Journal of Quality Technology" Vol 17, No 3 July 1985).

I have taken and continue to take a LOT of stat.

 

[xonik]

Yes, I didn't mean to insinuate that only the Antec's results were insignificant. That's one big thing I have against this review as a whole; I think I expressed these concerns in another recent thread.

 

[AACI_CIO]

Not really for you, xonik, since you seem to have some understanding, but a general message:

Statistical validity is a rare event on many websites. Who can blame the reviewers? Afterall, replication of a multiple hour/day task. Then again, if the website is your life, you should follow appropriate replication practives to boost validity. Review sites need replication in one of the following (if not both) forms:

1) Within unit replication (repeated measures on the same unit to make inferences about that unit).

2) Across units replication so that statisitcal inference may be made back to the population of units (arguably not needed if quality control of the compoenents is high--the components are all very similar).

Not many review sites follow either principle of scientific study.... As noted before, the time consumption is huge (I should know, I do large experiments every few months).

 

[Ice Czar]

snip snip

there are maybe 5 sites able to conduct a good PSU review
(or at least that Im aware of)
the rest are as mentioned, cheerleading squads happy to get freebies

SPCR is one; what are the others, Ice?

pcperspective
xbitlabs
systemcooling.com
amdmb and anandtech but its sort of hit and miss
your looking for individual reviewers that know WTF they are doing more than sites
in a few cases, the owner of the site and the reviewer are one and the same but not always
(note that Lee "Robotech" Garbutt, did the pcperspective, amdmb and the systemcooling reviews)

http://techreport.com/reviews/2004q4/psus/index.x?pg=9

What're your thoughts on this review?

somewhat flawed but better than most
http://techreport.com/reviews/2003q3/psus/index.x?pg=7
at least they are employing an oscilloscope, but they arent altering the AC input
or seriously giving the transient response a workout
temperature measurement is flawed all in all they get a C-
where most get D or F
their proceedure is still somewhat opaque so that might improve but thats my first take on it

Mike Chin and Lee "Robotech" Garbutt getting B+ > A-

I consider destructive and extreme dynamic load testing A+
If I win the lottery I'll do that and bust up a few mobos and supplies per week

did you review that anand tech memtest86 hack proceedure?
I thought that was very clever, and would like to see more of that

http://www.anandtech.com/casecooling/showdoc.aspx?i=1841&p=24

This is kind of a unique test we have worked on for a while. One of the more important qualities of a power supply is to make sure it does not produce harmful interference to the components. Perhaps the most sensitive of these components is the memory (video card, RAM, and CPU cache), which is just millions of little transistors locked in an on or off position, held in place only by a delicate balance of electricity.

The driving theory behind ECC memory is that it corrects errors that occur in memory. Perhaps a transistor is faulty and flipped for the wrong reason, or a faint electrical signal pulled a transistor into the incorrect position. While researching this review and others, we have noticed the largest factor for incorrect memory blocks is faulty power supplies.

How it works:

We started off by modifying MemTest86. We bumped the delay between tests up to several hours, rather than seconds. MemTests86 works by writing various patterns into the memory. After our programmed delay of six hours, the program goes back over the memory and analyzes how many of those bits were flipped out of the original pattern. Both the memory and motherboard used passed the unmodified MemTest86 without any problems.

Not only is this a good measure of shielding on the PSU, but also a good measure of shielding on the motherboard and memory. However, we used the same motherboard and memory for the entire test. This way we assure that the small amount of interference coming off the motherboard is consistent with each power supply. Unfortunately, it took 6 hours to run this test and we ran it three times for each supply to assure accuracy. On the next page is a table showing the trial data.

 

[Kain]

Glad I asked about that. Maybe cheezies'll do a good review of it.

 

[FreiDOg]

In defense of the Antec unit, 11.66 volts is within spec. Also, a single measurement of 11.66 volts is statistically insignificant.

Voltage tests were conducted over a 100 second time interval with a Pico ADC-212/3 digital oscilloscope. There was very little voltage deviation over the time interval, so the results are averages.

Now, testing a single unit doesn't give statistical relevance, as we well know, all the final output is only as accurate as the parts, each of them with their own tolerances.